KAOS, is a goal-oriented software requirements capturing approach in requirements engineering. It is a specific Goal modeling method; another is i*. It allows for requirements to be calculated from goal diagrams.[1] KAOS stands for Knowledge Acquisition in automated specification or Keep All Objectives Satisfied. The University of Oregon and the University of Louvain (Belgium) designed the KAOS methodology in 1990 by Axel van Lamsweerde and others. It is now widely taught worldwide at the university level for capturing software requirements.

1. Goal Modeling
KAOS

2. Goal-Directed Requirements
Acquisition (KAOS)
 (Organizational) goals lead to requirements.
 Goals justify and explain requirements which are
not necessarily comprehensible by stakeholders.
 Goals can be used to assign responsibilities to
agents so that prescribed constraints can be met.
 Goals provide basic information for detecting and
resolving conflicts that arise from multiple
viewpoints

3. The KAOS Modeling Philosophy
• Modeling a social setting involves a variety of concepts,
including Goals, Agents, Concerned Objects, Actions,
Constraints and Responsibilities.
• Goals lead to assignments of responsibilities and designs of
actions and artifacts
• Use a metamodel to support reuse of generic domain
modeling patterns [Dardenne93]
• For example, the library domain is an instance of the
“resource allocation” meta-domain, which also covers
car/room/dwelling rental and is similar to airline/hotel
reservation, class registration etc.

4. The KAOS Acquisition Process
• Identify Goals, and their Concerned Objects.
• Identify potential Agents and their Capabilities.
• Operationalize Goals into Constraints.
• Refine Objects and Actions.
• Derive strengthened Objects and Actions to Ensure
• Constraints.
• Identify alternative Responsibilities.
• Assign Actions to responsible Agents.
All this could be just as useful to someone doing
organizational design, rather than software development!

5. Four KAOS models :
• Goal model
• Responsability model
• Operation model
• Object model

6. Goals are desired system properties that
have been expressed by some stakeholders.
Stakeholder (wikipedia): third party who temporarily holds
money or property; more recent meaning (widely used in
management): person or organization that has a legitimate
interest in a project or entity; the new use of the term arose
together with and due to the spread of
corporate social responsibility ideas…
 interviews of domain experts and current & future
users
 analysis of existing systems
 reading of available technical documents
Goal modelGoal model

7. 7
• Each goal is either a root or is justified by at least
another goal that explain WHY it is needed.
• Each goal except a leave is refined as a collection of
subgoals describing HOW it can be reached.
• Identifying goals proceeds from either top-down,
bottom-up and non-directional approaches: from
intermediate goals are identified higher-level (business
or strategical) goals and lower-level (system
requirements) ones.
Goal modelGoal model

8. 8
Clear distinction between two low-level types of
goals:
• Requirements, to be achieved by a software agent,
• Expectations, to be achieved by an agent which is
part of the system environment.
Conflicts (and more generally obstacles that prevent
goals to be achieved) must be identified as soon as
possible.
Goal modelGoal model

9. 9
Domain properties:
Properties relevant to the application domain, used in
refinements to prove that a refinement is complete. Ex:
in order to satisfy a service request, an infrastructure to
perform the service must be available.
There are two types of domain properties :
- domain hypotheses: domain object properties expected
to hold ; ex: the elevator system has at least one cage to
carry passengers,
- domain invariants: properties known to hold in every
state of some domain object, e.g. a physical law,
regulation or a constraint enforced by some
environmental agent ; eg : for light buttons, we could
state that « the light is either on or off ».
Goal modelGoal model
Infrastructure available

10. 10
Requirements can be gathered by means of open
interviews. A more efficient way is to conduct less
open interviews be reusing requirements patterns.
One of the long-term benefits of investing in KAOS
technology consists in progressively modelling generic
patterns of requirements.
Completeness of the goal model is ensured be
reviewing the different diagrams during validation
sessions with stakeholders.
Goal modelGoal model

11. 11
• Information system: software system to be and the
part of the environment with which it interacts.
• Agent: humain being or automated component that
are responsible for achieving requirements and
expectations.
– We may stop refining goals into subgoals when they are
placed (as requirements) under the responsibility of a single
agent.
• Assignment is used to link expectations to related
agents (several agents may be assigned to an
expectation).
Responsibility modelResponsibility model

12. 12
Graphical Conventions
Goal & Responsibility modelGoal & Responsibility model
Goal
Requirement
Expectation
Agent
goal refinement
goals conflict
responsibility
assignment
Domain property
(pink)
(red)
(yellow)
(yellow)
(blue)
(blue)

13. 13
Completeness criterion 1: a goal model is said to be
complete with respect to the refinement relationship
‘if and only if’ every leaf goal is either an expectation, a
domain property or a requirement.
Completeness criterion 2: a goal model is complete
with respect to the responsability relationship ‘if and
only if’ every requirement is placed under the
responsability of one and only one agent (either
explicitly or implicitly).
Goal modelGoal model

14. 14
Goal modelGoal model
…with this generic goal pattern
(case driven decomposition)…
Beginning the Elevator case study…

15. 15
Goal modelGoal model...applied to the elevator problem:

16. 16
Goal modelGoal modelAnother generic goal pattern
(milestone driven decomposition)
Also appears
in following
schemes

17. 17
Goal modelGoal modelGeneric goal « Safe system »

18. 18
Goal modelGoal modelGeneric goal « Usable system »

19. 19
Goal modelGoal model
Generic goal « Service request satisfied »…

20. 20
Goal modelGoal modelGeneric pattern for the « Elevator called » goal
(milestone driven decomposition)

21. 21
Goal modelGoal model« Elevator called » goal

22. 22
Goal modelGoal model
How to deal with Alternatives:

23. 23
Goal modelGoal model
Qualitative comparaison:
• links to parent goals to which each requirement contributes
• creating a conflit if it contributes negatively
The design having a bidirectional button panel on each floor
seems to be the best compromise.

24. 24
Goal modelGoal model
Based on the bidirectional button panel design, we can
now refine the goal « Button depressed »:

25. 25
Goal modelGoal model
Let’s now refine the goal « Passengers brought to
requested destination », applying the milestone tactics:

26. 26
Goal modelGoal model
To refine the goal « No casualty », we use a generic
pattern that also use a milestone-driven refinement tactic:
Generic goal « No casualty »
The following figure shows how this pattern has been
used, and how the « No intrusion » goal has been
decomposed.

27. 27
Leaf goals are refined later:
•“Emergency stop available”.
•“System protected against fire”.
•“System protected against power failure”
•“No move in overweight conditions”.

28. 28
Goal modelGoal model

29. 29
Goal modelGoal model

30. 30
Goal modelGoal model

31. 31
Goal modelGoal model

32. 32
First shot of what an efficient elevator system should be:
Goal modelGoal model

33. 33
Responsibility modelResponsibility model
« System protected against fire » with responsibilities:

34. 34
After all requirements are assigned to a
responsible agent, a diagram is generated
for each agent.

35. 35

36. 36
Object modelObject model
The Object model is used to define and document the
concepts of the application domain that are relevant
with respect to the known requirements and to provide
static constraints on the operational system that will
satisfy the requirements.
Part of the Object model are objects pertaining to the
stakeholder’s domain and other objects for expressing
requirements or constraints on the operational model.
There are three types of objects: entities, agents,
relationships.

37. 37
 Entities: independant (needn’t refer to other objects)
and passive (can’t perform operations) objects; may
have attributes whose values define a set of states the
entity can transition to
 Agents: independant and active objects; they can
perform operations that usually imply state transitions
on entities (ex: elevator company, passenger, elevator
controler)
 Associations: dependant and passive objects that may
have attributes (ex: "At" that links a cage and a floor)
Object modelObject model

38. 38
Ex: entity « Alarm bell » as a component of the « Elevator »:
Object identification is driven by the goal definition process. Most
goals’s short and long definitions refer to domain objects being
modelled and documented.
All modelled objects shall have their own entries in the glossary
section of the requirements document. During review of the object
model, stakeholders will formally agree on a common vocabulary.
Another way to identify new objects consists in looking at the
requirements and discover the system components they are
necessary for satisfying them.
Object modelObject model

39. 39
Objects may be represented in goal diagrams; the concerns relationship
is used to link a requirement to the objects that are needed for it to be
satisfied. Identifying those objects shall further restrict the space of
solutions that can be proposed by the future system provider. Ex:
Object modelObject model

40. 40
Object « Elevator system »:
Object modelObject model

41. 41
Object « Cage »:
Object modelObject model

42. 42
Object « Floor »:
Object modelObject model
The KAOS model is compliant with UML class diagrams in that
KAOS entities correspond to UML classes; and KAOS associations
correspond to UML binary association links or n-ary association
classes. Inheritance is available to all types of objects (including
associations).
Objects can be qualified with attributes.

43. 43
The Operation model describes all the behaviors that agents
need to fulfill their requirements.
Behaviors are expressed in term of operations performed by
agents.
Ex: press button, enter, exit (perform by passengers) ; open
doors, close doors, move up, move down (perform by elevators).
Those operations work on objects: they can create objects,
trigger object state transitions and activate other operations (by
sending an event).
Operations are justified by the goal they operationalize: an
operation without a goal means there’re missing goals, while a
requirement without operations is just wishful thinking.
Operation modelOperation model

44. 44
Operations identification sources:
• stakeholders, that typically describe processes rather
than goals during interviews; the analyst then asks
specific questions in order to identify the reasons
behing the existing processes and hence unveil the
goals that justify these processes;
• existing requirements: operations explain how they
have to be realized.
Operation modelOperation model

45. 45
Requirements can be operationalized by:
• some object(s)
• some agent behavior(s)
• a combinaison of both
They are respectively requirements that describe
static, dynamic, and both properties on the system:
• the requirement « Elevator equiped with floor doors » will be
operationalized by an object « Floor door »
• requirements that describe dynamic system properties are
operationalized by operations
• the expectation « Stop button used » will be operationalized
with an operation « Push button » and the « Button » entity
Operation modelOperation model

46. 46
In a process, operations are represented as ovals,
concerned objects are connected to the operations by
means of input and output links.
Events are represented as: . Events may be
external or produced by operations (through an output
link). They may start (cause) or stop operations.
An operation model (or process) typically composes
operations performed by one or several agents to
achieve a requirement.
Operation modelOperation model

47. 47
Operation modelOperation model
Process
« Elevator control »

48. 48
Completeness criterion 3 : to be complete, a process
diagram must specify
(i) the agents who perform the operations
(ii) the input and output data for each operation.
Completeness criterion 4 : to be complete, a process
diagram must specify when operations are to be
executed.
Completeness criterion 5 : all operations are to be
justified by the existence of some requirements
(through the use of operationalization links).
Operation modelOperation model

49. 49
Operations can be triggered explicitely by an event. If
no event is specified, the operation will be triggered
implicitly when all the data needed in input are
available (data flow).
Requirements left unoperationalized are called « open
requirements »; solution providers have the complete
freedom to address the requirement as they want. The
less open requirements are left in the model, the more
precise the requirements document will be by
eliminating possible implementations.
Operation modelOperation model

50. 50
Partial classification of the events of the process:
Operation modelOperation model

51. 51
Typical instance of a process pattern KAOS analysts have
to reproduce in all projects:
Operation modelOperation model
(is responsible for)(operationalizes)
A requirement is said to be "closed" when this trian-
gular relationship Responsability-Operationalization-
Performance has been established.

52. 52
Operationalization diagram for the Reschedule operation:
Operation modelOperation model

53. 53
Dealing with obstacles.
Obstacles are situations in which a goal, a requirement
or an expectation is violated. The obstacle is said to
"obstruct" the goal, requirement or expectation.
Dealing with obstacles is very important for safety-
critical systems: it allows analysts to identify and address
circumstances at requirements engineering time
(instead of at programming or running time) in order to
produce for instance robust requirements or new
requirements to avoid or reduce impacts of obstacles.
The result will be a more reliable software.
ObstaclesObstacles

54. 54
Obstacle analysis with KAOS is a goal-oriented activity. It
begins with exploring the goal model and with negating
each goal in turn.
As a goal G is refined into goals G1, …, Gn if and only if
G1, …, Gn all together imply G, we know that if goal G is
violated, it is because at least one of its subgoals is
violated.
Each leaf obstacle is then reviewed with domain experts
to study whether it is worth considering it in the
obstacle analysis. It allows the analyst to prune the
obstacle space and to focus on the most relevant
obstacles.
ObstaclesObstacles

55. 55
Application to:
ObstaclesObstacles
casuality occurs during system use
some system component breaks down
somebody succeeds in hacking into the system.

56. 56
Other application:
ObstaclesObstacles
casuality occurs at service start
casuality occurs during service
casuality occurs at service end.

57. 57
Next, conditions for the obstacle to be raised are looked for. Typical
ways used to identify obstacles start by considering component
failures or people behaving in an unexpected way, maliciously or
because of some people’s capability limitation (disable people,
children, …).
Ex: an obstacle to the goal "Elevator direction updated few seconds
before next stop" occurs when blind people want to use the elevator
system as they can’t read the indications.
Obstacles are linked to the goals they obstruct (obstruction link).
Obstacles are refined the same way as goals, but while goals are
generally ‘AND-refined’, obstacles are most often ‘OR-refined’.
ObstaclesObstacles

58. 58
Obstacle to « Passengers informed of elevator direction »:
ObstaclesObstacles
OR refinement
(Obstruct)

59. 59
Having identified the conditions for it to occur, the
analyst can fix the obstacle in several ways. A first way
consists in defining new requirements that prevent the
obstacle from occuring.
In the preceding example, one can add a new
requirement: "Elevator direction announced by voice
upon floor arrival".
New requirements are linked to the obstacle they
resolve by a resolution link.
Other requirements for blind people could be "Floor
level announced by voice upon floor arrival" and "Call
buttons in Braille".
ObstaclesObstacles

60. 60
Obstacle resolution for « Passengers informed of elevator
direction »:
ObstaclesObstacles
(Resolve)

61. 61
An important source of obstacles concerns system (and
components) reliability and robustness. Reliability and
robustness requirements aim at preventing failures to
occur. One can for instance introduce requirements in
terms of Mean Time Before Failure (MTBF), or in terms
of a maintenance program to replace components
systematically at the right time, and so on.
A second way to fix an obstacle is to restore the
obstructed goal once the obstacle occurs. For instance, if
an elevator failure puts the system out of order, a
restoration requirement can prescribe the maximum
delay within which the system must be repaired.
ObstaclesObstacles

62. 62
• A third way to fix an obstacle consists in minimizing its
effects.
– Suppose the building is equiped with two elevator cages,
one serving floors 1 to 10, the other one floors 10 to 20. If
one cage goes out of order, a requirement should
prescribe that the system shall be self-reconfigured so that
the other cage provides service to all floors (from 1 to 20).
• It’s also possible to fix obstacles by modifying the
KAOS model in different ways, for instance by
substituting an agent with a more "capable" one, by
weakening the obstructed goal so that the obstacle is
no longer relevant or by substituting a goal to another
one.
ObstaclesObstacles

63. 63
Possible refinement of the goal « Robust and reliable elevator
system »:
ObstaclesObstacles
Each branch can be
opted for separately to
satisfy the goal, but
KAOS does not force
exclusion

64. 64
The structure of the requirements document generated
from a KAOS model is borrowed from the IEEE 830-1998
standard for software requirements specification :
Table of contents
1. Introduction
1.1 Document purpose
1.2 System purpose
1.3 Definitions, acronyms, and abbreviations → glossary derived from
1.4 References the KAOS model
1.5 Overview
2. Overall description
2.1 System perspective
2.2 User requirements → a subsection for each goal diagram
2.3 User characteristics and a list of the requirements
2.4 Constraints
2.5 Assumptions and dependencies → assumptions & obstacles not
2.6 Apportioning of requirements dealed with
ConclusionConclusion

65. 65
3. System requirements
3.1 System architecture → decomposition of the system by
KAOS agents (responsability model)
3.2 Object model → diagrams of the KAOS object model
3.3 Operation model → diagrams of the KAOS operation model
A major benefit of KAOS is that it provides a bidirectional traceability
between the problem description and the expected solution description.
As systems require frequent changes, it is vital to keep the requirements
document up to date with respect to the current state of development.
Requirements documents with KAOS tend to be more complete.
Completness must be understood by looking at the five completness
criteria: a complete KAOS model leaves no space for wishful thinking (a
goal not refined), requirements with no responsible, unjustified
operations, and operations for which we ignore who will execute what
and when.
ConclusionConclusion

66. 66
Completeness of the goal model also relies on 3 points :
 the quality of the requirements definition phase
(interviewing stakeholders, if operated properly, leads to
a good goal covering rate)
 the validation meetings during which stakeholders
review consolidated goal diagrams (conflicts are
addressed, forgotten goals have a new chance to come
up)
 refinement techniques that guarantee completness
and consistency (generic refinement patterns that
require formal notations for safety-critical applications).
ConclusionConclusion